// CCA Exam Reference

CCA Quick Reference
Cheatsheet

Independent Claude architecture study notes. One page. Zero fluff.

5 domains, 10 anti-patterns, 6 scenarios. Key concepts, code patterns, and the study topics — all in one reference card. This is not an official exam blueprint or a pass guarantee. Click domains to expand, search to filter, or print for offline study.

API notes checked September 27, 2026. Consult the Agent SDK, Claude Code memory, and structured output documentation for your installed version. Percentages below are study priorities, not a verified current exam blueprint.

// Section 01

Study Overview

Study Domains

5

Architecture through reliability

Pitfalls

10

Review with working examples

Study Scenarios

6

Practice architectural tradeoffs

// Section 02

Suggested Study Emphasis

D1 Agentic27%
D2 Tools/MCP18%
D3 Claude Code20%
D4 Prompts20%
D5 Context15%
// Section 03

Key Concepts

|
stop_reason: "tool_use" requests tool execution; "end_turn" ends this model turn. Also handle max_tokens, refusals, and other stop reasons.[Lesson 123]
Hub-and-spoke: route work through a coordinator; other communication patterns are possible.[Lesson 124]
Check context inheritance for your SDK/version: fresh subagents and forks differ. Pass task-specific evidence explicitly.[Lesson 124]
Current Agent SDK examples use allowedTools: ["Agent"] for delegation; older versions used "Task".
Anti-pattern: Natural language parsing for agentic loop termination
Agent = LLM + tools + loop. Without the loop, it is just a chatbot.
Agentic loop: prompt -> LLM -> tool_use -> execute -> append result -> repeat
Orchestration patterns: routing (one-shot), pipeline (sequential), hub-and-spoke (coordinator)
// Section 04

The 10 Anti-Patterns

1
Tool OverloadD2

Enabling irrelevant or overlapping tools without measuring selection accuracy

2
NL Loop TerminationD1

Parsing natural language to decide when the agentic loop should stop

3
Context InheritanceD1

Assuming subagents automatically inherit the coordinator's context

4
Same-Session Self-ReviewD4

Treating same-session self-review as sufficient independent verification

5
Confidence-Based EscalationD5

Using self-reported confidence scores to decide when to escalate to a human

6
Progressive Summarization DriftD5

Repeatedly summarizing context until critical details are lost

7
Generic Tool NamesD2

Naming tools 'do_thing' or 'process' instead of specific, differentiated names

8
Missing Stop ConditionsD1

No explicit termination criteria — agent loops indefinitely

9
Raw Exception ForwardingD2

Passing raw stack traces as tool errors instead of structured isError responses

10
Monolithic PromptsD3

Cramming all instructions into one massive system prompt instead of layering CLAUDE.md files

// Section 05

Six Practice Scenarios

4 of 6 scenarios are randomly selected each exam. Prepare for all of them.

1

SaaS Multi-Agent Customer Support

D1D2
2

CI/CD Code Review Pipeline

D3D4
3

Research & Report Generation

D1D5
4

Data Pipeline Orchestration

D2D5
5

Enterprise Knowledge Base Agent

D4D5
6

Autonomous DevOps Agent

D1D3
// The Code Whisperer

academy.jeremyknox.ai

Explore related Academy lessons from fundamentals to architecture. Free and member tracks are available; lesson access varies.

Related Tools

Exam Prep

CCA Practice Exam

20 scenario-based questions across all 5 CCA domains.

Open tool
Reference

Prompt Pattern Library

12 production prompt patterns with copy-paste templates.

Open tool